Maintainer identified the version; fake now exposes only our two consumed reads and rejects a wrong argument before the parser runs.

Where do its response examples come from? A strict call signature can still return a made-up shape that agrees perfectly with a mistaken parser.

Normal and failed examples now come from that version's documentation; malformed cases are separately marked deliberate test inputs, not claimed controller responses.

Exercise the report after a good read followed by failure as well as failure on first launch, because stale values are easy to miss when each test starts with an empty screen.

Good-then-failed case caught the report retaining the old value under a fresh timestamp; corrected, with old observation and current failure separate. Reopening saved reports is still untested.

Keep that reopening case visible. The fake has already exposed a real display error, but it still hasn't checked saved-report loading or a connection to the installed controller.